Anthropic's prompt caching really should be better known. A lot of features that distinguish LLM vendors are incremental nice-to-haves, but prompt caching enables algorithms otherwise too slow and costly to consider:

→ View original post on X — @goodside